>A while ago someone posted a historical comparison of
>final poll results with actual election returns. The
>results seemed to indicate that the polling was fairly
>accurate. However, I'm wondering what that tells us
>about the accuracy of tracking polls. Perhaps
>pollsters use larger sample sizes or more rigorous
>methodology as the election approaches, but smaller
>samples and quicker questions in earlier "tracking"
>polls. Has anyone seen anything about this?
Tracking polls usually use pretty small samples (several hundred rather than 1,000+), and average several days' results to compensate for that. There's no difference in the sample size of polls taken just before an vote - they're just more accurate because they're closer to election day. Also, as Ruy Teixeira's been arguing, the likely voter screen doesn't make sense this far from an election, because likelihood itself may change from day to day. It should only be relevant in the final polls.
Doug